Posts by BenWilson
Last ←Newer Page 1 2 3 4 5 Older→ First
-
Hard News: Gower Speaks, in reply to Sacha,
Wow. Can you explain those a bit please.
I can tell you what they are, but I'm only guessing how to use them. They're scatterplots, each plotting one party against one other party. The dots are individual polls and the x val is one of the parties, the y val is the other. The middle diagonal has no graphs because you'd be plotting the party's performance against itself, and get a dead straight diagonal line of slope 1. So they're used to conveniently label the graphs. If you want to know which parties one of the graphs uses, look along the row for the y-axis party, and the column for the x-axis.
There's two graphs for each pair. This is actually somewhat redundant, because they show exactly the same information, the axes are just reversed. So row 1 col 2 shows National on the x axis and Labour on the y axis, but row 2 col 1 shows National on the y and Labour on the x.
Scatterplots are good for seeing if there's some kind of relationship. Does one seem to go up when the other goes up? Or down? The first type is probably a possible correlation, the second a negative. How tightly it's clustered to that line indicates how extreme the correlation is. So a sharp line will get close to 1 (which is what you'd get on a Labour/Labour plot - because the variables are obviously perfectly correlated to themselves). A big fuzzy oblong will get closer to zero.
-
Hard News: Gower Speaks, in reply to JonathanM,
Do the Greens steal votes from Labour? Does NZ First steal votes from National?
We're looking for pattern in those, right? To my eye, I see Labour/National negatively correlated. Labour/Green barely correlated. Labour/NZF hardly correlated. National/Green negatively correlated. National/NZF negatively correlated. And then the really strange one: Green/NZF positively correlated.
I translate negatively correlated as "stealing votes from each other". But I'm not sure how to translate positive correlation. They make each other get more??? I think it's probably because they both get strong rises going into the elections. Correlation!=causation there, although there is possibly a common cause.
Do you think you could put in the non-voter dependencies on those 4? Because a strong positive correlation anywhere is quite counterintuitive unless the votes come from the undecided pool. I'd be expecting to see NZF and Greens negatively correlated with non-voters.
-
Hard News: Gower Speaks, in reply to JonathanM,
Presumably these calculated percentages are based on the idea "if the election was held now". The only uncertainty is whether the polls have accurately captured what the population wants, and we get all our combined probabilities from that?
I ask because it obviously isn't going to be held now, so extrapolating the uncertainty of an electoral outcome some way into the future just as the uncertain in current polling data strikes me as quite bogus. That's like saying that the only uncertainty of the price of the NZD/USD at the time of the election is the uncertainty we have in it right now. In other words, we know exactly what it will be (which we obviously don't).
What I'm saying here is that the actual uncertainty calculations we're making about the polling data might be applicable right before the election, but at this point they're not. The polls will go up and down as they have over the last 6 months. The true variance in all the outcomes much be considerably more, the further out from the event we actually are.
Makes me think that modeling it as a random walk might make more sense, when making the long range predictions. I most certainly can't believe that the odds of a National government are currently 99.5%. If we had these polls the day before the election, then I'd believe it more.
-
Hard News: Gower Speaks, in reply to David Hood,
OK, so P(N|~W) is not 0.47 in those figures. It's 0.995. Which makes bmk correct.
-
Hard News: Gower Speaks, in reply to David Hood,
it just makes no judgement as to how likely that is within the 53%
That would be an awesome conditional probability to be able to get. Then we could work out the overall probability of National-led govt.
Just for fun since I'm swotting this stuff right now:
P(N) = P(N|W)P(W) + P(N|~W)P(~W) (Bayes)
where N="National led govt", W="Winston holds balance", and ~ means "complement of", and | means "given" and
P(W)=0.53
P(N|~W)=0.47
P(~W) = 1-P(W) = 0.47
Giving P(N)=0.53*P(N|W)+0.22So if P(N)=0.5, P(N|W)=(0.5-0.22)/0.53=0.53
Thus, if the original probabilities are correct, Winston only needs to be more than 53% likely to choose National, for National to have a better than even chance of winning the election. But even if Winston has a 90% chance of going with them (so P(N|W)=0.9), National still only has a 0.53*0.9+0.22=0.70 = 70% chance of winning the election, and even if he's absolutely certain to go with them, the best chance they have is 75%.
-
Hard News: Gower Speaks, in reply to JonathanM,
a small chance (about 0.05%) of either labour-led
Labour-led meaning "not requiring (but possibly including) Winston" here? I ask because "Winston decides" surely includes an unknown chance of Labour led government?
-
Hard News: Gower Speaks, in reply to bmk,
While I would want him to go with Labour; I really don’t think he will. If it were just Labour then it would be a realistic possibility but he won’t get along with the Greens and he doesn’t want to be the third party in the coalition.
We'll see on that. He's a real wildcard. The two blocs will have to haggle with him about what he gets. It might depend on who can swallow the biggest Winston bribe.
But we're quite a ways out from the election. Support's going to move around some more. It's still anyone's game.
-
Feed: Saints Preserve, in reply to Rebecca Williams,
I’ll post a great recipe for chilli tomato jam later … promise.
Will hold you to it. I made a cut down version of your sauce recipe. I just cut every ingredient down to 1/4 original. Tomato sauce was homemade (I'm currently drowning in them). It made maybe 800ml. Very nice indeed.
-
Hard News: Gower Speaks, in reply to JonathanM,
I would love it if more stats reporting was Bayesian. It's the only part of stats that really makes sense to me, because it answers the question you really want answered - the chances of things happening. Finding some mean/CI and framing some godawful backwards hypothesis test, and then reporting the answers off an arbitrary "very high, high, good, fair..etc" just makes me switch off. When someone says high, I always want to know how high.
It also has the advantage that you can apply conditioning really easily to work out other interesting questions.
-
Feed: Saints Preserve, in reply to Richard Aston,
That’s how I restored a failing lemon tree , lemon trees it seems just loved to be pissed on. Same for orchids.
I tend to pile lawn clippings under citrus, too. They don't like root competition, and it's a regular addition of nutrients that you've got to put somewhere anyway. Maybe 5cm deep spread right out to the drip line, which is the rough extent of the roots.
Last ←Newer Page 1 … 269 270 271 272 273 … 1066 Older→ First